Lindsey Witmer Collins is Founder and CEO of WLCM.AI, an Inc. 5000 company building AI solutions for enterprises, and Scribbly Books.gettyWe pointed AI at the part of the job my team was best at, and it made them worse at it. So we pointed it at everything underneath instead.​Last spring, I handed one of my best people a tool that wrote cold outreach for her, and inside of a month, she was worse at her job and also, frankly, starting to hate it.She wasn't slower. Faster, if anything. And the emails and DMs were fine. Clean, warm-ish, personalized in the way software means when it says personalized: her name, their name, one detail scraped from somewhere.The tell was in her slumped shoulders, not to mention the numbers: replies dried up.I build software with AI all day long, for my own companies and for other people's, and I had done the most obvious possible thing. Here is a person who writes emails. Here is a machine that writes emails. I automated her out of her work entirely. I took the heartbeat out of the place it belonged.The Wrong HalfHer job, in the order it actually happens:Figure out who matters—which national organization, which brand or which affiliate has an audience that overlaps ours. Work out why they'd care, which is different from why we care. Find the human inside the org who owns that decision. Find out what they launched last quarter, who they just partnered with, what they said on a podcast in March. Find their contact information, collect the relevant URLs. Put outreach on the calendar.Then talk to a person.She's extraordinary at the last one. She is not especially interested in any of the rest, and I don't think she should have to be—those are two different jobs wearing the same title. But the research was eating most of her week, and the tool I gave her sped up not only this, which she was grateful for, but also the 10 minutes at the end, which made her want to run. I had automated the only part she was irreplaceable at, the place her beating heart belonged, and left her inside a machine to pull levers and press buttons, more like a cog than the people-person she is.The Command CenterI already knew how to solve this. We'd been doing it for developers for eight years.At WLCM, a senior engineer or PM does the thinking up front—scopes the work, breaks it down, sequences it, schedules it and puts it on a Kanban board. By the time the task reaches the developer responsible for delivering it, the board says what to build today and provides necessary context. Nobody on my dev team has ever started a Monday by figuring out what Monday is for. The model keeps them where they’re meant to be—doing the actual work—and it's a big reason why our product is good and my team is happy.So I built the relationship team the same thing, shaped for their work instead of ours. They open one screen. It says who to reach out to today and why today. What happened the last time we talked, and when. What their organization just announced. Three things that are true about them and useful to know. Who we said we'd circle back with. It runs overnight, on AI, doing research that would take a person hours and that no person would enjoy.The one thing it will not do is write the email. That box is empty on purpose, and it stays empty.The result is that her morning starts at the interesting part. She reads for four minutes, picks up the phone and is a person getting to know other people.I had the right idea—build them the systems and structure to keep their days humming—but what I lacked was restraint. I didn’t know where to stop.UnderneathNearly every company I watch adopt AI points it at the output. The writing, the talking, the deck, the face—the customer-facing layer. There are two reasons, and neither is strategy. It's the easiest place to point it, because there's a chat box and you paste into it. And it's the easiest thing to count. Emails sent is a number. Emails answered is a harder number and takes longer to get.The other version—AI underneath the person instead of in front of her—takes actual software. Somebody has to design it and build it and maintain it and rebuild it when she decides she'd rather see the follow-ups on the left. That used to be the entire obstacle. Bespoke internal tools for three people never penciled out. You build one CRM for 500 salespeople; you do not build three command centers for three humans who each want theirs different.That's what changed, and it’s one of my favorite things AI has done. It didn't only make writing cheap. It made software cheap enough to build one person at a time, which means the right answer is finally affordable. Every person gets a system built for the specific work they do and the specific way they like to do it. When they change their mind about how they work, you rebuild it. That's a Tuesday now.Which leaves me with an irony I've made my peace with: The most valuable thing AI has done for my company is make it cheap enough to build software that keeps my best people from ever having to open an AI.The machine eats whatever is routable, and in eating it makes us hungrier for the things it can't reach. A woman in Wisconsin who runs a manufacturing company we’d like to work with is not going to make her decision by talking to an AI we prompted. She's going to decide by getting to know us. And she’s going to get a call from a kind person who likes people, who has done her research and who has all morning.I make software. My team makes friends.​Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?